Classifications

• • G—PHYSICS
  • G06—COMPUTING; CALCULATING OR COUNTING
  • G06F—ELECTRIC DIGITAL DATA PROCESSING
  • G06F9/00—Arrangements for program control, e.g. control units
  • G06F9/06—Arrangements for program control, e.g. control units using stored programs, i.e. using an internal store of processing equipment to receive or retain programs
  • G06F9/46—Multiprogramming arrangements
  • G06F9/52—Program synchronisation; Mutual exclusion, e.g. by means of semaphores
• • G—PHYSICS
  • G06—COMPUTING; CALCULATING OR COUNTING
  • G06F—ELECTRIC DIGITAL DATA PROCESSING
  • G06F12/00—Accessing, addressing or allocating within memory systems or architectures
  • G06F12/02—Addressing or allocation; Relocation
  • G06F12/08—Addressing or allocation; Relocation in hierarchically structured memory systems, e.g. virtual memory systems
  • G06F12/0802—Addressing of a memory level in which the access to the desired data or data block requires associative addressing means, e.g. caches
  • G06F12/0844—Multiple simultaneous or quasi-simultaneous cache accessing
  • G06F12/0846—Cache with multiple tag or data arrays being simultaneously accessible
• • G—PHYSICS
  • G06—COMPUTING; CALCULATING OR COUNTING
  • G06F—ELECTRIC DIGITAL DATA PROCESSING
  • G06F12/00—Accessing, addressing or allocating within memory systems or architectures
  • G06F12/02—Addressing or allocation; Relocation
  • G06F12/08—Addressing or allocation; Relocation in hierarchically structured memory systems, e.g. virtual memory systems
  • G06F12/0802—Addressing of a memory level in which the access to the desired data or data block requires associative addressing means, e.g. caches
  • G06F12/0844—Multiple simultaneous or quasi-simultaneous cache accessing
  • G06F12/0855—Overlapped cache accessing, e.g. pipeline
  • G06F12/0857—Overlapped cache accessing, e.g. pipeline by multiple requestors
• • G—PHYSICS
  • G06—COMPUTING; CALCULATING OR COUNTING
  • G06F—ELECTRIC DIGITAL DATA PROCESSING
  • G06F12/00—Accessing, addressing or allocating within memory systems or architectures
  • G06F12/14—Protection against unauthorised use of memory or access to memory
  • G06F12/1458—Protection against unauthorised use of memory or access to memory by checking the subject access rights
• • G—PHYSICS
  • G06—COMPUTING; CALCULATING OR COUNTING
  • G06F—ELECTRIC DIGITAL DATA PROCESSING
  • G06F13/00—Interconnection of, or transfer of information or other signals between, memories, input/output devices or central processing units
  • G06F13/14—Handling requests for interconnection or transfer
  • G06F13/16—Handling requests for interconnection or transfer for access to memory bus
  • G06F13/18—Handling requests for interconnection or transfer for access to memory bus based on priority control
• • G—PHYSICS
  • G06—COMPUTING; CALCULATING OR COUNTING
  • G06F—ELECTRIC DIGITAL DATA PROCESSING
  • G06F15/00—Digital computers in general; Data processing equipment in general
  • G06F15/16—Combinations of two or more digital computers each having at least an arithmetic unit, a program unit and a register, e.g. for a simultaneous processing of several programs
  • G06F15/177—Initialisation or configuration control
• • G—PHYSICS
  • G06—COMPUTING; CALCULATING OR COUNTING
  • G06F—ELECTRIC DIGITAL DATA PROCESSING
  • G06F12/00—Accessing, addressing or allocating within memory systems or architectures
  • G06F12/02—Addressing or allocation; Relocation
  • G06F12/08—Addressing or allocation; Relocation in hierarchically structured memory systems, e.g. virtual memory systems
  • G06F12/0802—Addressing of a memory level in which the access to the desired data or data block requires associative addressing means, e.g. caches
  • G06F12/0806—Multiuser, multiprocessor or multiprocessing cache systems
  • G06F12/084—Multiuser, multiprocessor or multiprocessing cache systems with a shared cache

Definitions

• This invention relates to a digital data processing system, and more specifically, to a multiprocessor system
• - ⁇ 1 H _ comprised of a plurality of digital data processor ' s sharing common control circuitry, memory and peripherals.
• multiprocessor systems A number of digital data processing systems have been built heretofore based on multiple processors. These systems, referred to as multiprocessor systems, have had many diverse constructions. They have ranged from independent processors with no interconnection, through tightly coupled processor networks that communicate with each other by passing messages, to multiprocessors that. share common control circuitry, memory and peripherals. An excellent discussion of multiprocessor systems and of the trade-offs involved in system designs depending upon the degree of coupling and sharing involved in the systems is presented in Chapter 16 of the book entitled Computer Engineering, by C. Gordon Bell et al , Digital Press (1978).
• multiprocessor systems comprised thereof have the potential of providing computational capabilities that span ranges typically covered by many conventional single processor systems at comparable costs. For the end user, they offer the advantage of enabling modular incremental growth over limited and well-defined performance ranges with minimum training and other field service costs.
• Another object of the invention is to provide a multiprocessor system that exhibits performance capabilities comparable to conventional single processor systems at comparable cost.
• Another object of the invention is to provide a multiprocessor system that takes advantage of proven technologies of memory management and input/output control.
• Still another object of the invention is to provide . a multiprocessor system that exhibits a high degree of multiprocessing efficiency while maintaining a compatibility with existing processor protocols.
• An objective of the this invention is to provide a multiprocessing system that can be constructed of a varied number of modular components including processor modules, memory modules, mass storage devices, and input/output devices which together intercommunicate through and share a common control unit.
• a more specific objective of this invention is to provide such a modular multiprocessing system, the processors of which, share a common cache memory without undue degradation of overall system performance.
• a more general objective of this invention is to provide such a modular multiprocessing system that can be
• O PI economically assembled in one configuration, and subsequently expanded into a larger configuration as the demand for processing and data storage capacity increases without the necessity of replacing or modifying existing system components.
• a further objective of this invention is to provide such a modular multiprocessing system wherein the common control unit is compatible with various synchronous and asynchronous protocols, address translating circuits, and memory management routines desired in multiprocessing systems having different operating characteristics.
• a multiprocessing system comprises a common control unit including a memory management circuit, an address translation circuit, and a cache memory which shared in an ordered pipelined sequence by at least one processor, at least one mass storage controller including at least one data storage device, at least one memory module, and at least one input/output device (e.g., an I/O terminal). All processors connect to the common control unit via a non-interlocked synchronous backplane processor bus.
• the processor bus comprises a set of address and control conductors for transferring with the common control unit device identification signals, memory address signals and function code signals; and a set of data and control conductors for carrying processed, storage, and memory data together with other control signals.
• All input/output devices connect to the common control unit via an asynchronous input/output bus whose signals are also transferred to the common control unit via the processor bus.
• Memory modules also connect to the common control unit via an asynchronous memory bus.
• All . mass storage controllers connect to the common control unit via a synchronous mass storage bus. The memory bus and the mass storage bus connect to a memory interface circuit in the common control unit for accessing main
• the processors and input/output devices when necessary, also access main memory through the memory interface circuit.
• commands can be initiated on the processor bus, mass storage bus, or the input/output bus thereby to transfer information with any responding device connected to the processor bus, memory bus, mass storage bus, or the input/output bus, as appropriate.
• the common control includes control means so that it may access its own elements to update or allocate its cache memory and/or return memory data to the commanding device.
• the preferred structure of the common control unit comprises a separate control information path and a separate storage data path for respectively transferring among the devices of the multiprocessing system, device identification, memory address signals, and function code signals, on one hand; and processed data, storage data, memory data, and associated control signals, on the other hand.
• control information is transferred through a multi-stage register, the first of which receives device identification signals, function code signals, and virtual address signals. Internal control signals and physical address signals are generated by the common control unit, and then are recirculated in control information registers for transactions involving a second pass through the common control unit.
• the control information path also includes, at one stage, a memory management circuit that restricts the type of operations that can be performed with certain locations in the memory modules; at a subsequent stage, an address translation circuit that converts virtual addresses
• OHPI - - generated by a commanding device into a physical address in a responding device; and at a yet subsequent stage, a cache compare circuit controls read, update, and allocate operations in a cache memory.
• the data path of the common control unit also includes a multi-stage register arrangement.
• a first- stage register receives processed or memory data from a commanding device; or processed, storge, or memory data retrieved from a responder device.
• Another aspect . of the invention includes a Processor Index RAM and a Write-in-Progress flag. Because certain transactions initiated by a commanding device require a- second pass through the resources of the common control unit to update the cache memory on a "write hit" or allocate the cache memory on a "read miss", and because another commanding device may alter the contents of cache memory before a second pass completes the transaction, the cache memory further includes a Processor Index Random-Access-Memory (PIR) having an index location associated with " each processor for storing an address tag of a location being written and an associated write-in- progress (WIP) flag, which when set, inhibits other commanding devices from reading that location in the cache memory. When the write is completed on the second pass through the common control, the WIP flag is cleared.
• PIR Processor Index Random-Access-Memory
• the multiprocess system includes, preferably in the commanding device, a WIP flag control bit that is used for inhibiting a commanding device from issuing the subsequent write command while its -first write command is in progress.
• the common control unit can accommodate a plurality of identical processors, memory modules, mass storage devices, and input/output devices and thereby
• OMPI provide an economical expendable modular multiprocessing system.
• Fig. 1 is a block diagram of a multiprocessing system embodied in accordance with this invention
• Figs. 2A through 2E depict various clocking signals that synchronize transfers of information between the elements shown in Fig. 1;
• Fig ⁇ ._.3A through 3E depict the timing intervals for the elements shown in Fig. 1 when a processor performs a read operation to a memory location contained in the cache of Fig. 1;
• Figs. 4A through 4D depict the timing intervals for the elements shown in Fig. 1 when a processor performs a transaction requiring two accesses to the cache of Fig. 1;
• Fig. 5 r is a table listing internal operations performed in the common control unit in response to various conditions occurring therein;
• Fig. 6 depicts a more detailed diagram of the common control unit used in Fig. 1;
• Fig. 7 depicts a diagram of the internal memory arbitration and memory return circuit of the common control unit of Fig. 6;
• Figs. 8A and 8B show functional elements of the shared resources in the common control unit of Fig. 1. Description of Illustrative Embodiment
• information means to include device identification signals, memory address signals, signals that control the operation of a device (e.g., control and timing signals), and data signals from memory, storage, or processing devices.
• information broadly includes address, control, and data signals.
• device means any component of a data processing system.
• memory data refers to information associated with a random access memory module.
• storage data means information associated with a mass storage device, such as a disk file drum unit, or magnetic tape unit.
• cache data means information associated with a cache memory.
• data means memory data, storage data, register data, or cache data.
• the expression “storage information” includes data and its concommitant control signals which are routed to/from a data storage device.
• identification signal or “ID signals” means signals that identify a device in a data processing system.
• address signal or “memory address signal” means signals that identify an addressable location in a data storage device.
• function code or “control bits” means signals that specify how a device is to function or operate.
• control information includes address, control and data signals which together determine how a device will function or operate within a given set of operations.
• bus means a plurality of parallel conductors to which several devices may connect in parallel for the transfer of information.
• connection means the electrical connection between a device and a bus.
• the expression “connection” means a direct interconnection between two devices by way of a signal conductor or plural conductors.
• a "commanding” device is a device which issues a command.
• a “responding” device is a device which responds to a commanding device.
• a “transaction” means all the operations necessary to perform a command initiated by a commanding device. In this specification, it means the performance of all steps in an order pipeline sequence of events that are necessary to complete the command.
• a multiprocessing system constructed in accordance with this invention.
• the CCU 10 includes what are termed “shared pipeline resources” including a cache memory 20, a memory management circuit 22, an address “translation circuit 24, and a clocking and control circuit 26.
• shared pipeline resources include input registers for receiving sets of command signals and output registers for receiving data signals associated with the sets of command signals.
• the cache memory 20 is used for storing copies of information contained in random access memory modules 28 so that data can be quickly accessed by any one of the processors 30 (P, P, . .
• the memory access time for retrieving data might, for example, be 500 nanoseconds while the rate with which the CCU 10 can make a memory reference to the cache memory 20 is 133 nanoseconds.
• a cache memory circuit 20 that may be used with this invention is described in incorporated U.S. Patent No. 4,055,851.
• the memory management circuit 22 in the CCU 10 permits appropriate accesses to certain read-only and write-only, locations in the random access memory modules 28 according to preesta ' bli ⁇ hed rules " .
• the memory management " circuit 22 also provides a relocation constant that is used by the address translation circuit 24.
• a memory management circuit 22 which could be used in this invention is that described in incorporated U.S. Patent No. 3,893,084 issued July 1, 1975.
• the address translation circuit 24 is used for converting virtual addresses to physical addresses.
• Virtual addresses are memory addresses generated by a commanding device that do not define a physical address which uniquely defines a specific address location in a responder device.
• Physical addresses define unique addresses.
• the translation circuit 24 establishes a physical address by using an arithmetic unit to add the relocation constant from a register in the circuit 22 to the virtual address obtained from the commanding device.
• the address translation circuit 24 converts an 18-bit addresses generated by a processor 30 into 22 bit addresses that uniquely define a unique location in memory modules 28.
• Such an address translation circuit 24 that can be used with this invention is also described in incorporated U.S. Patent No. 3,854,126 issued December 10, 1974.
• the common control unit 10 also includes a memory request queue 34, subsequently described, for queing requests for access to the memory modules 28.
• the CCU causes the memory request queue 34 to store ID signals identifying the commanding, device, physical address signal identifying the physical address in memory modules 28 involved in the transfer, and function code signals (e.g. read or write command) identifying the operation that is to be performed with the memory-modules 28.
• These requests are then transmitted to a memory interface circuit 36 via a connection 38.
• the queue 34 holds 32 sets of command information.
• a queue is necessary because the request demand varies during ⁇ Operation of the multiprocessing system. These requests may emanate from control circuits in the clocking and control circuit 26, from processors 30, I/O devices 32, or the cache memory circuit 20.
• the memory interface circuit 36 also receives requests to retrieve or store storage information from the mass storage controllers 40 and 42 through a bus connection 44.
• the controllers 40 and 42 are connected to a 32-bit wide synchronous mass storage bus 46 which transfers both storage and control information.
• the mass storage but 46 and controllers 40 and 42 are described in U.S. Patent No. 3,999,163 and is commercially known as ⁇ a PDP 11/70 Large Scale memory bus.
• a conventional arbitration circuit in the memory interface 36 determines whether requests from the memory request queue 34 or the mass storage controllers 40 or 42 will be honored. Although not shown, the memory interface circuit could also be adapted to handle requests for access directly from the UNIBUS devices 32 via UNIBUS interface circuit 64. Once arbitration is completed, both control information signals and data signals are transferred to a memory bus 48, which signals are then transferred to an appropriate one of the memory modules 50, 52, or 54.
• the memory bus 48 also contains other control circuits for routing control information and memory data between the memory modules 28 and the memory interface circuit 36. A more detailed description of the memory bus 48 and memory modules 28, with associated control logic, is described in incorporated U.S. Patent 4,045,781.
• a memory return queue 65 stores the addressed memory data together with ID signals and the physical address from whence it came. Data to be.written into the ' emory modules 28 also is stored in the queue 65. Write data is used for a subsequent update of the cache memory 20. This information is transmitted back to the CCU 10 to be recirculated in the pipeline resources to update or allocate the cache memory 20, if necessary, and ultimately transferred to the commanding device that originally requested the memory data. Whether or not a cache memory update is made depends upon the result of address comparisons in the duplicate tag store 67. Its advantage and operation is subsequently described.
• Transfers of control information and data also take place between I/O devices 32, such as peripheral device 56 and terminal 58 (e.g., operator console, 0 teletypewriter or even a remote processor), via an asynchronous I/O bus 60, commercially known as a UNIBUS and described in incorporated U.S. Patent No. 3,710,324.
• I/O devices 32 access the common control unit 10 through the P-Bus interface 64. 5 These commands first enter the UNIBUS interface circuit 64 via a bus connection 62 and subsequently are transferred to the P-Bus interface circuit 61 via separate address and data bus connections 66.
• clocking and control circuit 26 controls the ⁇ execution of the commands originating from the I/O devices 32 to access internal registers in the pipeline resources, the cache memory 20, the memory modules 28, or the mass storage controllers 40 and 42 or their devices 68 and 70.
• processors 30 communicate with the common control unit 10 via a processor bus 76, also called a P- Bus.
• processors 30 may comprise any number of processor units 78 through 84. • Each of the processors 30 connect to the P-Bus 76 through respective interface circuits 86 through 92. These circuits include distributed arbitration networks for
• Each interface circuit 86 through 92 couples the P-Bus 76 via their respective pairs of address and data bus connections 93 through 100.
• the P-Bus 76 is a non-interlocked synchronous backplane bus and includes a separate "address" bus 102, and a separate "data" bus 104. Each of the buses 102 and 104 include a number of parallel conductors.
• Address bus 102 includes conductors for carrying device identification information (ID) signals, function code signals (F) signals, memory address information (ADRS) signals, and other control information, such as parity checking bits and timing signals.
• ID signal identifies a commanding device when directed to the CCU 10. Otherwise, the ID signal identifies the responding device.
• Data and control bus 104 carries data signals between the processors 30 and cache memory 20. The processors 30 may also communicate with each other via the P-Bus 76 without using the pipeline resources 20 through 26.
• the respective interface circuits 86 through 92 include a set of "address" bus connections 93,
• bus connections 93 through 100 enable the processors 30 to communicate with the common control unit 10 via separate address bus conductors 106, on one hand, and separate data bus conductors 108, on the other hand.
• Each set of conductors establishes a bus connection between the P-Bus 76 and P-Bus interface 61.
• the separation of conductors permit the simultaneous use of bus 102 by one device, and use of bus 104 by another device. For example, if processor 78 is performing a read command that does not require use of the data bus 104 of the P-Bus 76, the common control unit 10 need not wait to use bus 104 for returning data to an awaiting U
• processor or UNIBUS device This arrangement improves overall performance of the multiprocessing system.
• Operations carried out in the P-Bus and in the processes 30 are synchronized with an internal clock of the CCU 10. Specifically, an internal four-phase clock in the clocking and control circuit 26 drives the operations on the P-Bus 76 and processors 30.
• the processors 78 through 84 are PDP 11 data processing systems manufactured by the assignee hereof. These systems are described in incorporated U.S. Patent No. 3,614,741. Their description also may be obtained from the "Micro-computer Processor Handbook" (1979). published by the assignee thereof. Because the transactions occurring in the pipeline resources are non-interlock, certain other transactions can occur in the common control unit 10 prior to the cache memory being allocated or updated with * the new information. Interlocking of commands that alter the contents of cache memory.20 would degrade overall system performance.
• a Processor Index Ram (PIR) 20C that obviates the necessity to interlock these transactions and yet maintain time-wise efficiency in a multiprocessing system.
• PIR 20C Processor Index Ram
• the cache memory circuit 20 of the common control unit 10 is a direct mapping, write- through, cache memory. That is, the random access memory modules 28 are always written before the cache 20 can be updated. Therefore, new data written in the random access memory modules 28 must be reflected in the cache memory 20, if the physical address being written corresponds with one of the addresses in tag store 20A. Otherwise, cache memory data could become stale. Since we provide a write-through feature, certain operational requirements must exist among the devices of the
• the cache memory 20 contains a data store 20B having a certain number of address storage locations, and an address tag store 20A having a corresponding number of data storage locations.
• the data store 20B contains copies of memory data located in the memory modules 28.
• the specific address location in memory modules 28 that are duplicated in the data store 20B is identified by the corresponding address information contained in a corresponding location in tag store 20A.
• the cache memory 20 is said to be “associated” with the memory modules 28, and therefore, it is also referred to as an "associative" memory.
• the pipeline resources of the CCU 10 checks the tag store 20A to determine whether or not it must access memory 28 to retrieve the requested data. If negative, cache data is returned to the commanding device from the cache memory 20B at the end of the first pass through the pipeline resources. Thus, in this case, the commanding device, on its request for information from the memory modules 28, is said to "hit” the cache. Oppositely, the commanding device is said to "miss” the cache when the requested data is not formed therein. In this case, any cache data appearing on the data lines is invalidated at the end of the first pass through the pipeline resources.
• the clocking and control circuit 26 queues the request in the memory queue 34 for a subsequent request to the memory modules 28.
• the transaction initiated by the commanding device cannot be completed until a second pass is made through the pipeline resources.
• the second pass is made by the CCU 10 on behalf of the commanding device to read the data. When the requested data is retrieved,- it is then returned to the commanding device
• the commanding device awaits the requested information.
• the common control unit 10 allocates both the tag store 20A and data stored 20B with the new information from the memory modules 28.
• the write-through feature requires that the system embody a write-in-progress (WIP) -control bit in WIP circuits 79, 81, 83 and 85 associated with processors 78, 80, 82 and 84, respectively; and a Processor Index RAM (PIR) associated with the cache 20.
• WIP write-in-progress
• PIR Processor Index RAM
• This control bit is cleared by the CCU 10 on its second pass through the pipeline resources to update itself in response to the previously issued "write hit" issued by the processor. Since the PIR only contains one address storage location for each processor, each processor 30 checks its WIP control flag bit 79, 81, 83, and 85, respectively, and if set, foregoes the issuance of its subsequent write command.
• DIR Device-Index-RAM
• WIP WIP control flag bit
• the common control unit 10 checks a duplicate-tag store 67, subsequently described, to determine if it contains the address in memory to which the write command is directed. If affirmative, control circuit 26 updates the affected location in cache store 20B, while the corresponding tag store location remains unchanged. The update operation on this write cycle is interlocked so that no other operation can occur within the cache circuit 20 while being updated.
• the cache memory 20 is a direct mapping write-through cache memory.
• clocking and central circuit 26 controls a pieplined sequence of events in the CCU 10 with order preserved for all commands entering the pipeline.
• a four-phase clocking signal having 133 monosecond intervals, successively activates each resource in the common control unit in an ordered sequence thereby to step each command through the stages of the CCU 10.
• commands can be initiated by any one of the processors 30, UNIBUS devices 56 or 58, or mass storage controllers 40 or 42.
• the common control unit 10 also includes internal registers which can be read by a device to acquire status information. These control registers can also be written by any device or with control information generated by the CCU itself thereby to control its internal operation when it makes a second pass to update or allocate its cache memory and to pass the requested data to a commanding device.
• the P-bus is divided into two sections: an address and control section 102, and a data and control section 104.
• the address section carries virtual address signals (VA) which identify memory locations, function code signals (F) which indicate what type of transaction is being done, and identification signals (ID) which tell what devices are involved in the transaction. For example, the ID signals may tell what device initiated the transaction or to what device the transaction is directed. In those cases where the ID identifies the initiating device, the transaction is directed to the common control CC.
• VA virtual address signals
• F function code signals
• ID identification signals
• Access to the P-bus 76 is controlled by arbitration circuits that are distributed among the respective devices connected to the P-bus 76 (i.e., the common control CC and each of the processors PC) .
• These arbitration circuits which form part of the P-bus interface circuits 86, 88, 90, 92, and 61, are described in the copending application of John V. Levy et al. Serial No. 954,456 which is assigned to the assignee hereof and incorporated herein. Each such device has a distinct priority and an independent request line that ⁇ forms part of the address section of the P-bus.
• the device To gain access, to the P-bus, the device asserts its request line, and if no higher priority request is present when the P- bus becomes available, the device wins control of the P- bus.
• the common control CC is assigned the highest priority.
• the processors PC have distinct, lower priorities determined by their respective positions along the P-bus. Requests for. access to the P-bus from the UNIBUS 60 are serviced through the request line of the common control CC.
• a clocking circuit 26 in the common control CC distributes the timing signals that control transactions over the P-bus. These timing signals are distributed to each of the interface circuits 86, 88, 90, 92, and 61. As shown in Figs. 2A through 2D, the timing signals include P0 CL through P3 CLK quadrature-phase signals, each having a 50% duty cycle. The leading edges of the PO CLK through P3 CLK pulses define P0 through P3 clocking times, as shown in Fig. 2E. The time interval between successive P0 clocking times defines a complete bus cycle and therefore is called a bus cycle time. In general, any device seeking access ' to the P-bus asserts its request line at P0 during a given bus cycle. Arbitration is performed between P0 and P3 of that same bus cycle, and if control is won, it is awarded at P3 of that cycle. Other operations follow sequentially, as described below.
• the interval between successive P0 times, and therefore, the bus cycle time is about 133 nanoseconds.
• the P-bus 76, and those parts of the common control unit 10 needed by a processor Pn to read data from the 0 cache 20, are organized in a pipeline. Further, the design of the pipeline is optimized to minimize the access time of a processor Pn doing a memory read that hits in the cache 20.
• the resources in the pipeline are the following: 5
• any device wins control of the P-bus it is allocated all of the above pipeline resources. There is no way for any device to request only part of the resources in the pipeline. However, not all of the 0 resources need be used by the device. Thus, one or more resources may be idle during a given transaction if those resources are not necessary for the transaction.
• a device that wins control of the P-bus uses the address section of the P-bus during a first time slot that starts immediately when the P-bus is allocated. The device then
• Fig. 3 illustrates the timing for a processor Pn doing a memory read that hits in the cache 20.
• the clocking times P0 through P3 of Fig. 2E are repeated in Fig. 3A.
• processor PC0 desiring to perform a memory read, asserts its request line at P0 of a given bus cycle. If the processor P has the highest priority of those requesting the P-bus at that time, it is awarded the P-bus at P3 of the same bus cycle.
• the time interval between the request and the grant is labeled REQUEST in Fig. 3B.
• the processor P uses the address section of the P- bus to transmit a function code F indicative of a memory read, its ID, and the virtual address VA of the memory location to be read to the common control CC- This time interval is labeled TRANSMIT in Fig. 3B.
• the processor P uses the memory management circuit 22 in the common control CC. During the MANAGE interval, the memory management circuit checks the function code F and the virtual address VA received from the processor PC0 to determine if the location defined by the address is one for which a read
• the circuit 22 If the location can be read, the circuit 22 also generates a relocation constant.
• the processor P uses the address translation circuit 24 in the common control unit 10.
• An arithmetic unit in the circuit 24 adds the relocation constant to the virtual address VA received from the processor PCO to generate a physical address PA.
• the next interval is CACHE.
• the CACHE interval extends a full bus cycle, that is, from the last P0 until the P0 that follows it, and is thus the longest time interval of all the resources in the pipeline.
• the CACHE interval is made long enough to permit only one access to the cache 20.
• the next interval is HIT CHECK. During this interval,' the cache hit check circuit in the common control unit 10 is used to determine if the read hit in the cache 20.
• the HIT CHECK interval extends from the P0 at the end of the CACHE interval to the next P3.
• the last interval in Fig. 3B is the DATA interval.
• the data section of the P-bus is used to transmit the data read from the cache 20 back to the processorP .
• the memory read is completed.
• the hit check circuit indicates that the memory read hit in the cache 20
• the ID sent by the processor P is not needed.
• the processor P which is synchronized to the P-bus timing, strobes the data section of the P-bus during the DATA interval to receive the data.
• the ID is needed when the memory read misses the cache 20, as explained below.
• Tm,n refer to a transaction time that occurs on the n-th clocking time after the -th bus cycle after the beginning of a transaction.
• the processor PCO described above doing a memory read asserts its request line at T0,0, uses the address section of the P-bus from TO,3 until Tl,2, and so on until it receives the data at T5,2.
• the timing of the pipeline for the processor PCO is shown in Fig. 3B in terms of transaction times Tm,n.
• Fig. 3D shows the timing for a read hit that is initiated by a processor P having a lower priority than the processor P described above.
• Fig. 3E shows the timing for the read hit by the processor P in terms of transaction times Tm,n.
• Fig. 6 illustrates further details of the common control CC.
• the operation of the common control unit 10 is best illustrated by describing how it responds to memory reads that miss in the cache 20 and to memory writes. Read misses will be discussed first.
• processor P desires to read data from a memory location that is not contained in the cache 20.
• the processor P will proceed synchronously through the resources of the pipeline in the same manner described above for the processor P doing a read that hits in the cache 20.
• data is sent back to the processor P through the data registers of the P-bus during the DATA interval of the pipeline even if the hit check circuit 162 determines that the read missed in the cache 20. The data returned is thus incorrect.
• the hit check circuit 162 upon detecting a miss, negates a DATA VALID signal and transmits the negated DATA VALID from the register 163 signal back to the processor P over the data section of the P-bus with the incorrect data.
• the negated DATA VALID signal puts the processor P in a WAIT state until, during TRANSMIT interval of a • subsequent pipeline sequence, the processor P senses its own ID on the P-bus. The correct data for the location read by the processor P is received four cycles later from the data section of the P-bus during the DATA interval of that subsequent pipeline sequence.
• Transaction times are indicated above the components shown in Fig. 4. They indicate the time when the corresponding component completes its operation in the pipeline.
• the processor P is attempting a read from a memory location that is not. contained in the cache 20.
• the hit check circuit in the cache 20 enters a function code F for a read, the ID of the processor P, and the physical address PA of the memory location to be read in the registers of multiplexer 158.-
• T6,l the same information is entered in the appropriate queues or buffer 172, 174, or 176. Once the information is stored in the queues or buffer, the first pass of the read miss through the pipeline resources is completed.
• the memory bus 48 operates asynchronously with respect to the P-bus.
• the memory references are invoked asynchronously through a memory bus arbitration circuit.186 until the queue is empty.
• the memory bus arbitration circuit 76 is described in the above referenced U.S. Patent No. 4,045,781. Order is ' preserved in each of the queues 172 and 176 and therefore, all memory accesses are thus treated on a first-in-first-out basis.
• a command wins control of the memory bus 48 it accesses the desired location in the backing store memory units 184.
• the physical address PA and data read from the memory units 184 are passed from the memory bus 48 to a memory return queue 188.
• the memory return queue 188 also operates on a first-in-first-out basis.
• a duplicate tag store 182 is also maintained at the entry to the memory return queue 188 as a copy of the tag store 20A in the cache 20. Upon entry of the data in the memory return queue 188, a duplicate tag in the duplicate tag store 182 is written indicative of the physical address PA of the memory location just read. The purpose of the duplicate tag store 182 will be explained below.
• the common control CC With the data from memory stored in the memory return queue 188, the common control CC must request the pipeline in order to again access the cache 20 and to send the correct data back to the processor Pn, which initiated the read miss.
• the common control unit 10 thus initiates a new transaction by asserting its request line at T0,0 of a new bus cycle.
• the common control CC must arbitrate for access to the P-bus.
• the common control CC has the highest priority of all P-bus devices.
• the common control CC places a function code F indicative of a read, the physical address PA and the ID from the memory return queue 188 on the address section of the P- bus.
• the common control CC allocates the cache 20 by writing the tag store 20A of the cache 20 with the physical address PA of the memory location just read and by inserting the data in the corresponding location of the cache data store 20B.
• the data is placed on the data section of the P-bus for reception by the processor P. The read miss operation is then complete.
• write operations to locations in memory are simple.
• the device initiating such a write reguests the pipeline, places virtual address information VA on the address section of the P-- bus during the TRANSMIT interval, places the data to be written on the data section of the P-bus during the DATA interval, and is then done.
• For the common control CC however, writes are more involved.
• the common control CC must update the cache 20 if the location being written is in the cache 20. Since the order of the resources in the pipeline is such tha.t the data to be written is not available to the common control CC when the cache 20 is accessed during the first pass through the pipeline, a second pipeline pass must be initiated.
• the common control unit 10 detects a function code F indicative of a write to memory, it places the function code F, the physical address PA of the location to be written and the data to be written in the register 158. The information is then transfered to the queue 172 or 176 or to buffer 174, depending upon the ID signals, when the first pass through the pipeline resources is completed.
• the command wins control' of the memory bus 48 it does a write in the backing store memory units 184.
• the data from memory and the physical address PA of the location just written into memory 184 are loaded into the memory return queue 188.
• the duplicate tag store 182 is then checked to determine if the memory location just written is one that is contained in the cache 20. If it is not, the common control CC is done processing the write. If it is, then the common control CC requests the pipeline to initiate a second pass through the resources of the pipeline. During the second pass, the data store 20B of the cache 20 is updated with the data written into memory.
• the duplicate tag store 86 is maintained as a copy of the tag store 20A in the cache 20.
• the validity of this copy is maintained by interlocking changes in the duplicate tag 182 with changes in the cache'tag store 20A.
• the duplicate tag store 182 prevents a particular cache location that has yet to be updated by a write that hits in the cache 20 from being reallocated to different memory location between the two passes of the write through the resources of the pipeline. Such an erroneous reallocation could occur, for example, if the second pass of a read that missed the cache 20 occurred in time between the first and second pass of a write.
• cache location X corresponds to memory location A.
• the first pass of a read through the resources of the pipeline that misses the cache 20 is completed, but not the second pass of that read.
• a write to memory location A is received by the common control CC. Because memory location A is contained in cache location X, the write would be treated like a cache hit.
• the second pass of the read that missed the cache 20 allocates cache location X to a different memory location, e.g., memory location B.
• the second pass of the write being treated as a cache hit updates cache location X with the new data from memory location A. This results in erroneous data being stored in cache location X.
• This problem is avoided according to the invention through the implementation of the duplicate tag store 182 at the entry of the memory return queue 188. It was 0 recognized that only entries from the memory return queue 188 can change a cache tag. By placing the duplicate tag store 182 at the entry to the memory return queue 188 and by preserving order through the queue 188, it can be assured- that the tag found in the duplicate tag store 182 5 is identical to that actually in the cache 20 at the time the particular entry reached the cache 20. If a particular cache location X happens to be reallocated to a new memory location before the write completes its second pass through the pipeline, a check of the 0 duplicate tag store 182 reveals this.
• the duplicate tag store 182 serves another purpose.
• the contents of the backing store memory units 184 can be changed by the mass storage controllers 40 and 42. When a memory location that is in the cache 20 is changed by an I/O transaction from the units 40 or 42, " this change must be reflected in the contents of the cache 20. It is inefficient, however, to go through the cache 20 continually whenever the contents of the memory units 184 are changed by the units.40 or 42.
• transaction from the units 40 or 42 like those from P-bus devices, check the duplicate tag store 182 prior to initiating a cache update. If the duplicate tag store 182 indicates that the location written to by the units 40 or 42 is in the cache 20, the cache 20 is updated; otherwise, it is not. This allows the cache 20 to be used only when an actual update is necessary thereby eliminating continual use of the cache 20 for every memory location written by the units 40 or 42. In this manner, the cache 20 can be utilized for P-bus activity without sharing the bandwidth of the cache 20 with updates from the units 40 and 42.
• This problem could be avoided by invalidating the- location in the cache 20 that is being written on the first pass of the write through the resources of the pipeline. However, this would force the processor performing the write to be stalled for the entire duration of the write. It could also result in the invalidation of a cache location being used by another P- bus device. Both would degrade overall system performance.
• a Processor Indexed Ram (PIR) 167 is implemented as part of the cache 20.
• the PIR 167 is used .in lieu of any cache invalidation scheme.
• the PIR 167 includes one location for each processor 30 in the system 10.
• the PIR 90 is indexed by the ID of the processors PC and is one bit " wider than the 12-bit index of the cache 20. The additional bit in the PIR index is used as an indicator of when a write operation is in progress.
• the location of the PIR 167 corresponding to the processor Pn performing the write is written with the tag of the memory location to be written.
• the additional bit in that location of the PIR 167 is set.
• the read operation is treated as a read miss even if the location to be read is in the cache 20.
• the read is thus forced to enter the memory queue 172 behind the write. . If it is not, the read operation proceeds as it would normally, i.e., as a read hit if the
• Fig. 4B illustrates the timing for a read operation that is initiated by the processor P and that misses the cache 20.
• Fig. 4C illustrates the timing for a write operation that is initiated by a processor P that hits
• Processor P has a lower priority than processor P .
• Fig. 4D illustrates the timing for a write operation that is initiated by a processor P that misses the cache 20.
• Processor P has a . lower priority than processor P and processor P .
• the PIR 90 includes only one location for each processor PC in the system 10.
• the PIR 167 can account for only a single write in progress for each processor Pn . * It is possible, however, for any given processor Pn to have more than one write operation in progress at a given time. If that happens, the PIR 90 prevents only the first of the locations being written to from being read. The succeeding locations being written to can be read by the processor PC, resulting in the return of invalid data.
• this problem is avoide through the implementation of a WRITE IN PROGRESS (WIP) flag in the hardware of the processors 30.
• the WIP flag prevents any single processor Pn from having more than one write outstanding at any given time.
• a processor PC desires to initiate a write operation, it checks its WIP flag before it requests the P-bus. If the WIP flag is set, then the processor PC continues to check the WIP flag until the WIP flag is cleared. When the WIP flag is cleared, the processor Pn requests the P-bus, transmits its ID, a write function code F and the VA of the location to be written over the address section of the P-bus, and then sets its WIP flag.
• the PIR 167 could be implemented so as to include more than one location for each processor Pn in the system 10. The PIR 167 would then be capable of accounting for the case where a single processor PC has more than one write operation outstanding at any given time.
• input/output devices 32 (which might also include a remote processor) and the processors 30 initiate transactions with the pipeline resources by sending ID, function code, and virtual address signals to the high-speed P-Bus interface circuitry 61.
• the memory modules 28 are only accessed by the CCU 10 when these transactions "miss” the cache when performing a read command, or "hit” the cache when performing a write command. As previously explained, some of these transactions occur synchronously, others occur asynchronously, and some may even directly access the memory modules 28 or internal command/status registers of the CCU 10 when the command so authorizes.
• the mass.storage controllers 40 and 42 on the other hand, always directly access the memory modules 28 via the memory interface 36 and therefore do not use the resources of the pipeline.
• the common control unit 10 does monitor write commands issued by the mass storage controllers 40 and 42, and if a location in the cache memory 20 is affected, an update of the cache is made by the CCU 10.
• requests for access to the memory modules 28 occur at various random time periods.
• Figs. 6 and 7 a queuing and memory arbitration arrangment for receiving the requests for access to the memory modules 28 thereby to provide efficient data throughput for our multiprocessing system. All queues operate on a first-in-first-out (FIFO) basis.
• a multiplexer 158 selects the inputs to the register 152 according to the function code signals and the internally generated CCF control bits CCF2-CCF0 generaged by a CCF generator, subsequently described. These control bits are altered from all zeroes on a first pass through the pipeline, as depicted in Table I, in response to function code signals, ID signals, and the hit check circuit 162.
• the CCU 10 includes control circuits responsive to these controls bits to alter the • internal operation of the CCU 10 on its second pass through the pipeline.
• the internal arbitration network 150 determines whether the CCU 10 gains access to its resources to update or allocate its cache 166, a UNIBUS device gains access to perform a transaction, or whether the memory arrangement 168 gains access to return requested data. Mass bus devices do not compete for use of the resources.
• a read "miss" occurred on a first pass through the pipeline. This could occur when the requested data is not in the cache 20 or when the request is made to a location for which there is a write in progress. No data is returned over the P-Bus 76. Instead, at T, a multiplexer 170 sends the appropriate function code, physical address, and ID signals to a processor/UNIBUS queue 172, a UNIBUS/MEMORY buffer 174, or to a processor/memory queue 176. Likewise, on a write "hit”, instead of updating- the cache, the multiplexer 170 queues the same signals in the appropriate queue 170 or 172 or buffer 174. It also includes control circuitry to select the data signals associated with the function code, physical ⁇ address and ID signals. The data signals, as previously noted, follow the address signals by f6ur P-Bus cycles.
• the processor/UNIBUS queue 172 passes its command to a UNIBUS interface circuit 178 which handles the command
• queue 172 accommodates up to 16 sets of data, physical address, function code, and ID signals. It is loaded at the T time period as previously indicated.
• a UNIBUS device responds to a processor/UNIBUS command, it places its data, function code, and ID signals in a UNIBUS return buffer 178.
• the CCU arbitration circuit 150 accepts information from the UNIBUS return buffer 178, the CCF generator sends the appropriate control information to the processor/UNIBUS queue 172 thereby to permit another set of command information to enter the UNIBUS interface circuit 178.
• the UNIBUS/memory buffer 174 stores a single set of data, address, and control signals that emanate from a UNIBUS device.
• a command from a UNIBUS device enters the UNIBUS return buffer 178, and once it successfully gains access to the pipeline resources, it gets read data from the cache 166 via the processor/UNIBUS queue 172 if the data is resident in the cache 166. Otherwise, it places its read or write command information in the UNIBUS/memory buffer 174. If the UNIBUS command resulted in a "read miss", the CCU returns the read data to the processor/UNIBUS queue 172 on a subsequent pass through the pipeline.
• the CCU notes that transaction and updates its cache 166 accordingly.
• the CCU generator notes the status of the transaction as they are stepped through the stages pipeline thereby to generate- the appropriate control bits CCF2-CCF0 for controlling the internal operation of the CCU 10 at each stage thereof.
• a processor/memory queue 176 receives commands from the processors 30 that are directed to the memory modules 184. These commands include read, write, write/modify, or reads and writes to internal registers resident in the 3 ⁇ - ⁇ - .
• Commands from devices on the mass storage bus 46 enter the memory arbitration circuit 176 from a buffer in the interface circuit 177.
• the only occasion during which a memory access by a mass storage device requires the use of the pipeline resources is when a mass storage device performs a write to the memory 184 at a location that is maintained in the cache memory 166.
• This determination is made with information from a duplicate tag store 182. Since the duplicate tag memory
• a CCU generator produces control bits
• control bits are supplied to a decoder in the
• a conventional read-only-memory (ROM) might constitute a decoder 180.
• the ROM might, for example, be recorded with information so that the contents of the memory address identified by the ID and function code signals represents the CCF bits stored thereat. These CCF bits are then recirculated in the pipeline thereby to control its action during the second pass.
• Some specific actions performed by the CCU in its preferred embodiment during the second pass include: selecting inputs to the register 152 by controlling multiplexer 158; activating queues 172, 174, or 176 to queue a request; updating or allocating the cache 166; updating the PIR; and updating the duplicate tag store 182. Other operations also may be performed.
• the specific gating* and control circuitry resides in the skill of a person ordinarily skilled in the art, and therefore are not herein shown in detail. All references to memory 184 from the processors 30, UNIBUS devices 32, or mass bus devices 68 and 70 enter the memory arbitration circuit 186.
• the memory arbitration circuit 186 decides which of the buffer 174, queue 172, and queue 176 will gain access to the memory modules 184 to perform, for example, a memory read or write command.
• Read data returned from the memory module 184 is temporarily stored in a memory return queue 188 prior to being returned to the common control unit 10. Since the common control unit 10 employs a write-through cache memory 20, data that is written into the memory module 184 is also temporarily stored in memory return queue 188 for later use by the common control unit 10 to update its cache memory 20. Whether or not an update operation is performed depends upon whether a "write hit" occurs in the duplicate tag store 182.
• the duplicate tage store 182 keeps track of all allocations of the cache memory 20 so as to maintain a current list of address, tags corresponding to the memory data resident therein. * This feature permits the common control unit 10 to update its cache memory 20 on. a single pass through the pipeline resources. Two passes would otherwise be required because the hit check circuit is not utilized in the pipeline until after the cache memory 20 is accessed.
• Fig. 7 depicts the memory arbitration and return circuit 168.
• the memory bus arbitration circuit 186 receives physical address signals, function code signals, identification signals, and data signals (if a write command) from the processor/memory queue 172, from the UNIBUS/memory buffer 174, and from a mass storage interface circuit 177.
• Each of the connections comprises a separate set of conductors for transferring physical address signals, function code signals, identification signals, and data signals from a selected buffer 174, queue 176, or interface circuit 177.
• the drawings only reflect the paths along which these conductors extend and does not show the individual separate conductors. Selection can be performed by conventional techniques, such as rotational, assigned priority, or dependant upon number of commands in the queues.
• the physical address signals, write data (if any), function code signals, and ID signals are transferred over the conductors 198 to a command buffer 196. From the command buffer 196, these signals are conveyed to both the memory modules 184 and a memory command queue 200 over separate sets of conductors 202, 204, and 206.
• a command decoder 208 receives the function code signals and the ID. signals and decodes the same thereby to produce the appropriate control signals for controlling memory operations in the memory modules 184. If a read command is to be executed, memory data is returned from the modules 184 via memory bus connection
• O PI 210 to a multiplexer 212.
• control circuitry in the memory module 184 enters the write data at the location identified by the physical address on the physical address conductors 202, and then records the same write data in the memory command queue 200.
• multiplexer 212 is switched by the control logic in the memory command queue 200 in . order for transmitting read data from the memory modules 184 or for returning write data from the memory command queue 200 thereby to place the appropriate data, together with its function code, ID, and physical address signals, in the memory return queue 188, which is the preferred embodiment, can accommodate up to 16 sets of commands.
• the output of the memory return queue 188 is then conveyed to the mass storage bus interface circuitry 177, the common control internal arbitration network 150, and the duplicate tag store 182. If the function code and ID signals specify that read data is to be returned to a mass storage device, the. ass storage bus interface circuit 177 accepts the read data without use of the pipeline resources and therefore does not access, nor allocate, the cache memory 20 nor any other pipeline resource.
• the address of data written into memory modules 184 by a mass storage device, or by any other device for that matter, is monitored by a tag compare circuit 214, and if a write "hit" occurs in the duplicate tag memory 182, the CCU 10 requests access to the pipeline resources to update its cache memory 20. Such update operations are accomplished by transferring the write command from the memory return queue 188 to the internal arbitration network 150 which grants immediate access to the resources. Cache update commands have one of the highest priorities for access to the pipeline resources.
• duplicate tag store 1-82 contains addresses of all tag addresses of the cache memory 20 whether or not the cache tag store 20A has yet been allocated.
• Write data commands in the queue 188 that is scheduled to update the cahce memory 20 to access the duplicate tag store 182 after exiting from the queue 188.
• a write command will not update the cache if there is an allocate operation pending in the queue 188.
• the queuing arrangement prevents false updates of a cache location that will be imminently allocated by a read command resident, but not yet executed, in the queue 188.
• This arrangement obviates the need to interlock commands that alter the tag store 20A of the cache memory 20. * Such interlocking would seriously degrade overall performance of the system.
• the P-Bus 76 comprises separate address and data sections for conveying respectively address and control signals on one hand, and data signals on the other hand.
• the resource of the common control unit is divided into separate address section, depicted in Fig. 8A, and a separate data section depicted if Fig. 8B.
• the function code signals, ID signals, and address signals are placed in a first stage register 352 during a first P-Bus cycle, and the data signals are placed in the first stage register 306 of the data section of the common control unit four cycles later.
• the first set of command signals advances to the second stage registers 354, and the associated data signals with the second command, if any, is placed in the register 306.
• the previous set of data signals advanced to the second stage data register 308.
• each set of command information is placed in the register 352
• its corresponding data signal follows four cycles later in the data register 306.
• the advantage of transferring the data signals four cycles behind the function code, address and ID signals is apparent from a brief review of the common control unit 10 depicted in Fig. 6. Assume that a miss occurred in the cache 166 thereof. An access to the memory modules is made and the corresponding data is returned through the internal arbitration network 150.
• the function code, ID and physical address signals emanate from the arbitration network 150 onto data conductors 155 while the corresponding -data signals are carried via conductors 154 to the first stage register 152.
• the function code, ID, and physical address signals reach the P-Bus 76 by way of the output registers 163, the P-Bus device which initiated the read command is notified of its upcoming, data signals which will be propagated through the common control unit resources four cycles later. Since the function code, ID and address signals need not use the resources of the common control, specifically the memory management circuit and the address translation circuit 24, those resources can be available to other sets of command information that is being propagated through the pipeline resource. The separation of the address and data signals in this fashion provides for efficient operation and use of the common control resources.
• the CCF generator 340 controls the operation bf all the registers, multiplexers, and internal logic of the registers of Figs. 8A and 8B.
• the common control arbitrator 336 and the CCF generator 340 monitor the information signals associated with the commands as they are sequenced through the resources of the pipeline. On the first pass AD
• the CCF generator 340 receives information that is to be returned to the P-Bus 76 by way of conductors 370, 374, and 378. In response to that information, the CCF generator 340 produces a set of control bits CCF2-CCF0 which are then supplied to the CCF section of the register 360.
• Other control signals from the CCF generator 340 are supplied to a multiplexer control circuit 350 which controls the selection of the multiplexers 364, 366, and 368 on subsequent passes of command information through the pipeline resources. Additionally, the multiplexer control circuit 350 controls the selection of cache data on conductors 318 o pipeline data transmitted over conductor 316 through the multiplexer 314. The conditions by which the selection is made was previously explained.
• the multiplexer control circuit 350 also controls a multiplexer 300 ' thereby to select data signals from the pipeline via conductors 302 or data signals from the internal arbitration circuit 150 yia conductors 304.
• the common control unit 10 also includes a Processor Index RAM 326 shown in Fig. 8B.
• the Processor Index RAM contains an index position 325 for storing physical addresses and ID signals and a write-in-progress bit 327 for indicating whether the device identified by the ID signals therein has transmitted a write command that has not yet been written in the cache memory.
• the command information of a commanding device reaches the cache memory stage of the pipeline resources, it not onl checks the address tag store in the cache memory 328 to ascertain whether its requested data is resident in the cache memory, it also checks its index position 325 and its write-in-progress bit 327. Thus, even though the requested information is resident in the cache memory 328, if the write-in-progress bit 327 is set, the circuit 330 will force the commanding device to "miss" its memory reference to the cache.” The force "miss" circuit 330
• the common control arbitration network 336 allocates the return information to a P-Bus slot only if data is not being transmitted with another command ih the pipeline. If this type of time slot is available, the multiplexer control circuit 350 will cause the multiplexers 364, 366, and 368 to enter the function code signals, ID signals, and a portion of the physical address signals into the register 360, which signals are thereafter driven onto the P-Bus 76 at the same time that the requested read data is supplied to the pipeline input buffer 306.
• the common control unit When the data that was supplied to the data register 306 reaches the output buffer 320 just prior to being driven into the pipeline, the common control unit will have recirculated the function code signals, ID signals, and physical address signals to itself so that all of this information reaches the output buffers 360 and 320 at the same time so that they together may be - placed on the P-Bus 76.
• Other operations are also performed within the common control unit depending on the status of the CCF generator control bits CCF2-CCF0. These control bits are recirculated through the register stages 352, 354, 356, 358, and 350 and 360 thereby to initiate certain action within the processor index RAM 326 and the cache memory 328.
• Fig. 5 depicts all operations that are performed internally within the common control unit in accordance with the type of command, the physical address range, and the function 2
• code signals F3-F0 The appropriate information is supplied from the registers to control circuitry in the common control unit to effect these operations.

Landscapes

• Engineering & Computer Science (AREA)
• Theoretical Computer Science (AREA)
• Physics & Mathematics (AREA)
• General Engineering & Computer Science (AREA)
• General Physics & Mathematics (AREA)
• Software Systems (AREA)
• Computer Hardware Design (AREA)
• Computer Security & Cryptography (AREA)
• Memory System Of A Hierarchy Structure (AREA)
• Multi Processors (AREA)

Applications Claiming Priority (2)

Publications (3)

Family

ID=22365126

Family Applications (1)

Country Status (6)

Families Citing this family (50)

Citations (6)

Family Cites Families (14)

• 1980
  • 1980-01-28 US US06/116,083 patent/US4345309A/en not_active Expired - Lifetime
• 1981
  • 1981-01-28 EP EP81900540A patent/EP0046781B1/de not_active Expired
  • 1981-01-28 GB GB8127279A patent/GB2078408B/en not_active Expired
  • 1981-01-28 JP JP56500824A patent/JPS6319904B2/ja not_active Expired
  • 1981-01-28 DE DE813134428T patent/DE3134428T1/de active Granted
  • 1981-01-28 WO PCT/US1981/000126 patent/WO1981002210A1/en active IP Right Grant
• 1989
  • 1989-07-10 JP JP1177847A patent/JPH02168341A/ja active Granted

Patent Citations (6)

Also Published As

Similar Documents

Legal Events